Segmentation of skin lesions based on fuzzy classification of pixels and histogram thresholding

UTOMATED segmentation of skin lesions in dermoscopy images is currently a challenging problem [1]. This paper proposes an innovative method to address this problem developed by the authors. It has been structured as follows. Firstly, in this introduction, on the one hand the segmentation problem is described and, on the other, the evaluation criteria used (image database, ground truths and metrics) are shown. Secondly, the system design is presented. Thirdly, the results and the discussion are shown. A. Problems with segmentation of skin lesions in dermoscopy images Automated segmentation of a skin lesion is a complex issue, as the possible casuistry that can appear in the images is very diverse. The main problems that can de found in the image which make segmentation difficult are as follows: 1. Presence of hair; 2. Other artifacts such as electronic letters, rulers, ink and color charts, etc.; 3. Dark rectangular or circular marks around it (a consequence of shadow); 4. Flashes; 5. Lighting problems: apart from the problem with dark marks and flashes that have already been mentioned, in some cases one part of the image turns out to be darker than another (a common cases is that the part of the skin beside the circular marks is often darker as it is less brightly lit, and some images also turn out to be darker than others; 6. As a result of the oil used to acquire many images, there may be distortion problems and bubbles; 7. Presence of blood vessels; 8. Presence of regression areas and blue-whitish veil -in many cases these structures have greater intensity than the skin surrounding the lesion; 9. Hypopigmentation areas which are confused with skin; 10.